Research Of Semi-supervised Face Recognition By Convolutional Neural Networks Based On Graph Clustering

With the rapid development of computer network technology,traditional identification methods are facing severe challenges such as security and anti-counterfeiting.In order to ensure the security of personal information,people have paid their attention to biometric technology with good discriminative power,in which face recognition is favored for its convenience and directness.In recent years,with the in-depth study of deep learning,convolutional neural network has gradually become the mainstream method in the field of pattern recognition and image processing.However,convolutional neural network also has an obvious shortcoming,that is,the training of the network needs a large number of supervised data.But,in practice,it is difficult to obtain a large number of labeled data.In comparison,the acquisition of unlabeled data is very easy.Therefore,reviewing the principles of image preprocessing,face feature extraction,face cluster:ing and convolutional neural network,this paper proposes a face recognition method for processing semi-supervised data by combining face clustering with convolutional neural network.The main contents of this paper include:1?By means of piecewise Gamma transform,this paper proposes an improved face image data preprocessing method,which effectively improves the influence of illumination and image contrast on subsequent face detection2?In this paper,face clustering is used to process semi-supervised data.Current clustering methods such as k-means clustering and hierarchical clustering are difficult to process high-dimensional non-convex data.For this reason,this paper uses Chinese Whispers(CW)clustering algorithm based on graph theory to cluster human faces.This paper first introduces the direction gradient histogram(FHOG)and support vector machine(SVM)to achieve frontal face detection.At the same time,the Ensemble of Regression Trees is used to extract the key feature information of the face.Then use the CW algorithm to cluster.By comparison,the CW algorithm has achieved better results in both accuracy and other clustering evaluation indicators,successfully converting semi-supervised data into supervised data.3?To achieve face recognition through convolutional neural networks,this paper first designs three convolutional neural network models with different degrees of complexity.The clustered data and the combination of ReLU activation function and different pooling criteria are used to train the model and analyze the impact of different structures on network recognition capabilities.Then,this paper applies the dropout technology to the network to improve its generalization ability.Through comparative analysis,the network performance of adding the dropout is better than before.